The slide used in ITSC 2019. In this work we did followings:
- Simulate driver's attention on a 3D map
- Analyze the potential colliding hazard obstacles
- Show the boundary that divides expected and unexpected obstacles during the drving
Safety Criteria Analysis for Negotiating Blind Corners in Personal Mobility Vehicles Based on Driver’s Attention Simulation on 3D Map
1. Safety Criteria Analysis for Negotiating Blind
Corners in Personal Mobility Vehicles Based on
Driver’s Attention Simulation on 3D Map
Nagoya University, Japan
Naoki Akai, Takatsugu Hirayama, Luis Yoichi Morales, and Hiroshi Murase
2. Background
⚫ Can we say that autonomous navigation system are safe?
• There is a trade-off relationship between safety and speed [1]
• Over safe navigation compromises speed and smoothness
• How human drivers determine the trade-off relationship?
[1] Y. Yoshihara, L.Y. Morales, N. Akai et al. Autonomous predictive driving for blind intersections. In Proc. of the IEEE/RSJ IROS, pp. 3452-3459, 2017.
3. Motivation
⚫ Find the reasonable trade-off relationship from human's driving data
• Over safe is of course not suitable for autonomous navigation
• However, safety must be guaranteed
• Fina a point of compromise for autonomous navigation
Focus on wheelchair type
personal mobility vehicle (PMVs)
4. Approach
⚫ Show a limitation of human drivers with numerical values
• Humans have limitation, e.g., cannot observe obstacles locating at
occluded areas
• However, they can smoothly negotiate blind corners
• Assume that navigation under the limitation similar to human’s driving
⚫ Simulate driver’s attention on a 3D map using robotic technologies
• Analyze potential colliding hazard obstacles that drivers cannot
avoid if they rash out from occluded areas
5. Platform (Robotic wheelchair type PMV)
⚫ The PMV is able to estimate driver’s eye-gaze direction in a 3D map
• The PMV first recognizes its own position on the 3D map [2]
• The PMV then estimates the eye-gaze direction using a motion
capture that tracks the eye-gaze measurement glasses
• Occluded areas for a driver can be accurately estimated
[2] N. Akai et al. Mobile robot localization considering class of sensor observations. In Proc. of the IEEE/RSJ IROS, pp. 3159-3166, 2018.
6. Potential colliding hazard obstacles (PCHOs) simulation
⚫ PCHOs are obstacles that drivers cannot avoid if they suddenly rush
out from the occluded areas
• The minimum linear velocity and collision angle between the PMV
and the obstacle are recorded if there are PCHOs
Simulate obstacles that definitely
collide against the PMV
Unrealistic velocity and
collision angle are observed
Unexpected parameters for the
drivers are obtained
7. Example of driver’s attention and PCHOs’ simulations
⚫ The PCHOs (cubes) are observed when passing blind corners
• The color of the cubes represents level of the linear velocity
Movie
https://www.yout
ube.com/watch?
v=71jKnTve2-k
8. Experimental conditions
⚫ Driving of four participants were analyzed in an indoor environment
• One skill-full driver (SD) and three non-skill-full drivers (NSDs)
• One participant respectively drove three CW and CCW trials
[3] T. Hatada. Psychological and physiological analysis of stereoscopic vision. Journal of Robotics and Mechatronics, 4(1):13–19, 1992.
[4] T. Miura. Visual search in intersections: An underlying mechanism. IATSS Research, 16:42–49, 1992.
⚫ Perception ability of obstacles was
defined while referring [3, 4]
• 90 horizontal and 60 vertical degrees
• Assume that the drivers are able to
observe all obstacles which exist in
the field of view
9. Result by the SD (CW)
⚫ Example of a result by the SD in a blind corner
• Left: PMV’s trajectory with velocity and eye-gaze directions
• Right: PCHOs (size of the circles represents level of the velocity)
10. Results by the NSDs (CW)
⚫ Similar results to that of the SD were confirmed in the same corner
NSD1 NSD2 NSD3
Eye-gazedirectionsPCHOs
11. Comparison of eye-gaze behaviors
⚫ Eye-gaze angles of yaw and pitch axes are not similar
• The SD carefully watched left and right sides (top left)
• However, the similar PCHOs were observed in the all trials
12. Other results (CCW)
⚫ Similar results between the all participants were also confirmed
• The PCHOs were also found in the all trials
NSD1 NSD2 NSD3
Eye-gaze
directions
PCHOs
SD
13. Summarize the PCHOs’ parameters
⚫ The boundary can be seen in the collision angle-linear velocity plot
• The boundary could divide expected and unexpected obstacles
Expected
Unexpected
14. Conclusion
⚫ Motivation and approach
• Want to show numerical values to evaluate safe driving behaviors
• Develop the platform that is able to estimate the driver’s attention
in a 3D map, and simulate the PCHOs for the drivers
⚫ Results
• It was confirmed that the simulated PCHOs were unrealistic since
they have significant large velocity and collision angle
• It was shown that there is a boundary in the collision angle-linear
velocity plot of the simulated PCHOs
• We concluded that the boundary divides expected and unexpected
obstacles during driving, thus it could be the numerical criterion